An adaptive periodic background elimination method based on fourier analysis

Abstract

The present application relates to the technical field of image processing, and more particularly to a self-adaptive periodic background elimination method based on Fourier analysis, comprising the following steps: step one, image input; step two, horizontal and vertical frequency division; step three, respectively performing inverse color, enhancement and normalization preprocessing operations on the horizontal frequency graph and the vertical frequency graph; step four, performing Fourier transform on the image of step three, and performing Fourier transform on the horizontal frequency preprocessed graph and the vertical frequency preprocessed graph; step five, obtaining high-amplitude phase difference of the image of step four; step six, obtaining horizontal and vertical phases; step seven, high-response point elimination; step eight, inverse Fourier transform, performing inverse Fourier transform on the separated frequency response graph and phase response graph in step seven; and step nine, feature recognition; the algorithm of the present application can automatically identify the periodic grid background on the liquid crystal panel, automatically eliminate a number of frequency parts with the highest response in the horizontal and vertical directions, and the remaining non-periodic features after elimination are defects.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an adaptive periodic background elimination method based on Fourier analysis. BACKGROUND

[0002] In the production process of liquid crystal panels, defect identification on the liquid crystal grid is an essential part of the quality management process, and the defect map on the liquid crystal grid is shown in the accompanying drawings of the specification. Figure 1 The existing image recognition-based technology is divided into two categories: the first category is a template comparison technology based on traditional image algorithms, which compares the captured image with the ideal design drawing, and finally outputs the abnormal points after image correction, enhancement, matching, and identification of differences. The second category is a defect identification algorithm based on deep learning neural networks, which pre-establishes a training set for each type of defect. Based on the completion of a large amount of data training, the model has the ability to automatically identify defects, and the trained model can be deployed on the operation unit.

[0003] The disadvantage of the first category of template comparison technology based on traditional image algorithms is that design drawings need to be collected for each product, which is a tedious process. When a new product is put online, image algorithm personnel need to perform multiple parameter adjustments for the characteristics of the product, which reduces the efficiency of new product online.

[0004] The disadvantage of the second category of defect identification algorithm based on deep learning neural networks is that the model training period is long, the number of collected pictures is large, and the operation unit has high requirements. For defects that do not appear in the training set, multiple rounds of labeling and iterative training are required, which further reduces the deployment efficiency.

[0005] Therefore, the existing two categories of image recognition technology cannot meet the demand of improving image recognition efficiency. SUMMARY

[0006] The technical problem to be solved by the present application is to provide an adaptive periodic background elimination method based on Fourier analysis, which can effectively identify any type of liquid crystal panel, automatically extract the highest frequency part in the horizontal and vertical directions, and eliminate it in the original image. The remaining non-periodic features after elimination are defects, and the method can also automatically identify whether the defects invade the wire position and assist in judging the severity level of the defects.

[0007] To solve the above technical problems, the technical solution of the present application is: an adaptive periodic background elimination method based on Fourier analysis, comprising the following steps:

[0008] Step 1: Image input, input the original image with defects into the algorithm module;

[0009] Step two, horizontal and vertical frequency division, respectively, to obtain the image of each row and each column of the pixel average value, only horizontal frequency of horizontal frequency map and only vertical frequency of vertical frequency map;

[0010] Step three, respectively, to horizontal frequency map and vertical frequency map for anti-color, enhancement, normalization preprocessing operation, to obtain the horizontal frequency preprocessing map and vertical frequency preprocessing map;

[0011] Step four, Fourier transform of the image of step three, Fourier transform of the horizontal frequency preprocessing map, to obtain the horizontal frequency spectrum response map and horizontal frequency phase response map; Fourier transform of the vertical frequency preprocessing map, to obtain the vertical frequency spectrum response map and vertical frequency phase response map;

[0012] Step five, to obtain the high amplitude phase difference of step four image;

[0013] Step six, get horizontal and vertical phase;

[0014] Step seven, high response point rejection;

[0015] Step eight, inverse Fourier transform, Fourier transform of the separated spectrum response map and phase response map in step seven;

[0016] Step nine, feature recognition.

[0017] As a preferred technical solution, the preprocessing operation in step three, the anti-color preprocessing operation is to participate in the Fourier analysis of the important dark line as a positive signal; the enhancement preprocessing operation is to adjust the contrast of the image, so that the step of the positive signal is more obvious; the normalization preprocessing operation is to linearly scale the maximum value and the minimum value of the image to the top and bottom of the color gamut; the purpose of the above anti-color, enhancement, normalization preprocessing operation is to make the result after Fourier transform more clear and more easily identify the maximum response area.

[0018] As a preferred technical solution, the horizontal and vertical phase in step six, including obtaining the position of the maximum value of the horizontal frequency spectrum response map, and then obtaining the corresponding horizontal frequency phase response position, and then obtaining all the periodic characteristic information in the horizontal direction, that is, the horizontal frequency base frequency position; obtaining the position of the maximum value of the vertical frequency spectrum response map, and then obtaining the corresponding vertical frequency phase response position, and then obtaining all the periodic characteristic information in the vertical direction, that is, the vertical frequency base frequency position.

[0019] As a preferred technical scheme, the high response point elimination in step seven comprises performing two-dimensional bidirectional Fourier transform on the original image with defects to obtain an original image spectrum response graph and an original image phase response graph, and eliminating the base frequency positions corresponding to the horizontal frequency and the vertical frequency on the original image spectrum response graph and the original image phase response graph respectively to separate the periodic and non-periodic characteristics in the horizontal direction and the vertical direction.

[0020] As a preferred technical scheme, the feature recognition in step nine can clearly see the defects in the result after the periodic features are eliminated, and the subsequent foreground and background threshold segmentation or dynamic binary traditional image algorithm can extract the defect positions.

[0021] As a preferred technical scheme, in the Fourier transform result obtained by analyzing the horizontal and vertical frequencies in step four, the width of the periodic signal and the phase information of the periodic signal in the current image can be extracted; using the two kinds of information, the actual positions of the periodic background in the image can be marked, and the invasion position relationship between the defects and the liquid crystal grid can be identified by combining the defect positions and the actual positions of the periodic background, thereby assisting in judging the defect severity level.

[0022] Due to the adoption of the above technical scheme, an adaptive periodic background elimination method based on Fourier analysis comprises the following steps: step one, image input, inputting an original image with defects into an algorithm module; step two, horizontal and vertical frequency division, respectively calculating the pixel average value of each row and each column of the image to obtain a horizontal frequency graph with only horizontal frequency and a vertical frequency graph with only vertical frequency; step three, respectively performing inverse color, enhancement and normalization preprocessing operations on the horizontal frequency graph and the vertical frequency graph to obtain a horizontal frequency preprocessed graph and a vertical frequency preprocessed graph; step four, performing Fourier transform on the images in step three, performing Fourier transform on the horizontal frequency preprocessed graph to obtain a horizontal frequency spectrum response graph and a horizontal frequency phase response graph, and performing Fourier transform on the vertical frequency preprocessed graph to obtain a vertical frequency spectrum response graph and a vertical frequency phase response graph; step five, calculating the high amplitude phase difference of the images in step four; step six, obtaining the horizontal and vertical phases; step seven, high response point elimination; step eight, Fourier inverse transform, performing Fourier inverse transform on the separated spectrum response graph and phase response graph in step seven; and step nine, feature recognition; the algorithm of the present application can automatically identify the periodic grid background on the liquid crystal panel, and can effectively identify any type of liquid crystal panel. The identification algorithm can automatically extract the highest response frequency part in the horizontal and vertical directions and eliminate it in the original image, and the remaining non-periodic features after the elimination are defects. This method can also automatically identify whether the defects invade the wire position, thereby assisting in judging the severity level of the defects. BRIEF DESCRIPTION OF DRAWINGS

[0023] The following drawings are only intended to illustrate and explain the present application and do not limit the scope of the present application. Among them:

[0024] Figure 1 is a defective original pattern of an embodiment of the present application;

[0025] Figure 2 is a flowchart of an embodiment of the present application;

[0026] Figure 3 is a horizontal frequency-only pattern of an embodiment of the present application;

[0027] Figure 4 is a vertical frequency-only pattern of an embodiment of the present application;

[0028] Figure 5 is a horizontal frequency spectrum response pattern of an embodiment of the present application;

[0029] Figure 6 is a horizontal frequency phase response pattern of an embodiment of the present application;

[0030] Figure 7 is a vertical frequency spectrum response pattern of an embodiment of the present application;

[0031] Figure 8 is a vertical frequency phase response pattern of an embodiment of the present application;

[0032] Figure 9 is a pattern spectrum response pattern of an embodiment of the present application;

[0033] Figure 10 is a pattern phase response pattern of an embodiment of the present application;

[0034] Figure 11 is a periodic part pattern of an embodiment of the present application after inverse Fourier transform and elimination;

[0035] Figure 12 is a result pattern of an embodiment of the present application after elimination.

[0036] In the figure: 1 - defect; 2 - wire; 3 - liquid crystal grid. DETAILED DESCRIPTION

[0037] The present application will be further described with reference to the drawings and embodiments. In the following detailed description, certain exemplary embodiments of the application are described by way of illustration only. It will be readily apparent to one of ordinary skill in the art that numerous modifications to the described embodiments can be made without departing from the spirit and scope of the application. Accordingly, the drawings and description are to be regarded as illustrative in nature and not restrictive.

[0038] As Figures 1 to 12As shown in the drawings, an adaptive periodic background elimination method based on Fourier analysis comprises the following steps: Step 1, image input, inputting an original image with defect 1 into an algorithm module; the original image with defect 1 is as shown in Figure 1 As shown in the drawings, defect 1 is shown as a black dot in Figure 1 Step 2, horizontal and vertical frequency division, respectively obtaining the pixel average value of each row and each column of the image to obtain a horizontal frequency graph with only horizontal frequency and a vertical frequency graph with only vertical frequency, so as to effectively avoid the interference caused by horizontal and vertical frequency overlap; the horizontal frequency graph with only horizontal frequency is as shown in Figure 3 As shown in the drawings, the vertical frequency graph with only vertical frequency is as shown in Figure 4 Step 3, respectively performing inverse color, enhancement and normalization preprocessing operations on the horizontal frequency graph and the vertical frequency graph, a series of preprocessing operations are needed to be performed on the image before periodic analysis, so as to obtain a horizontal frequency preprocessed graph and a vertical frequency preprocessed graph; Step 4, performing Fourier transform on the image of Step 3, performing Fourier transform on the horizontal frequency preprocessed graph to obtain a horizontal frequency spectrum response graph and a horizontal frequency phase response graph; the horizontal frequency spectrum response graph is as shown in Figure 5 As shown in the drawings, the horizontal frequency phase response graph is as shown in Figure 6 As shown in the drawings, performing Fourier transform on the vertical frequency preprocessed graph to obtain a vertical frequency spectrum response graph and a vertical frequency phase response graph; the vertical frequency spectrum response graph is as shown in Figure 7 As shown in the drawings, the vertical frequency phase response graph is as shown in Figure 8 As shown in the drawings, since the input is a single horizontal or vertical frequency image, the output result is also only data distribution in the horizontal or vertical direction; Step 5, obtaining high-amplitude phase difference of the image of Step 4; Step 6, obtaining horizontal and vertical phases; Step 7, high-response point elimination; Step 8, Fourier inverse transform, performing Fourier inverse transform on the separated spectrum response graph and phase response graph in Step 7, wherein the eliminated periodic part containing only horizontal and vertical frequencies is as shown in Figure 11 As shown in the drawings, the result after elimination is as shown in Figure 12 Step 9, feature recognition; the beneficial effects of the present application are as follows: the algorithm of the present application can automatically identify the periodic grid background on the liquid crystal panel, and can effectively identify any type of liquid crystal panel. The identification algorithm can automatically extract a plurality of frequency parts with the highest response in the horizontal and vertical directions, and eliminate them in the original image. The remaining non-periodic features after elimination are defect 1. This method can also automatically identify whether defect 1 invades the position of lead 2, thereby assisting in judging the severity level of defect 1.

[0039] The pre-processing operation in step three, the inverse color pre-processing operation, is to take the important dark lines as positive signals to participate in Fourier analysis; the enhancement pre-processing operation is to adjust the contrast of the image to make the step of the positive signal more obvious; the normalization pre-processing operation is to linearly scale the maximum and minimum values of the image to the top and bottom of the color gamut; the purposes of the above inverse color, enhancement, and normalization pre-processing operations are to make the results after Fourier transform clearer and more easily identify the maximum response area. The inverse color, enhancement, and normalization pre-processing operations are all prior art in the field, and will not be described here.

[0040] The acquisition of the horizontal and vertical phases in step six includes obtaining the positions of a plurality of frequency maximum values of the horizontal frequency spectrum response graph, then obtaining the corresponding horizontal frequency phase response positions, and then obtaining all the periodic characteristic information in the horizontal direction, i.e., the horizontal frequency fundamental frequency position; obtaining the positions of a plurality of frequency maximum values of the vertical frequency spectrum response graph, then obtaining the corresponding vertical frequency phase response positions, and then obtaining all the periodic characteristic information in the vertical direction, i.e., the vertical frequency fundamental frequency position. The spectrum response positions here can be used for background periodicity removal, and the phase response positions can be used for defect position identification.

[0041] As shown in Figure 9 and Figure 10 , the high response point removal in step seven includes performing two-dimensional bidirectional Fourier transform on the original image with defect 1 to obtain an original image spectrum response graph and an original image phase response graph, as shown in Figure 9 and Figure 10 , respectively. The horizontal frequency and vertical frequency fundamental frequency positions are removed from the original image spectrum response graph and the original image phase response graph, respectively, to separate the periodic and non-periodic characteristics in the horizontal and vertical directions.

[0042] The feature identification in step nine, i.e., the result after removing the periodic characteristics, can clearly see the defect 1, as shown in Figure 12 . The white point in the image is the defect position, and the subsequent traditional image algorithms such as foreground and background threshold segmentation or dynamic binarization can extract the defect position. The traditional image algorithms such as foreground and background threshold segmentation and dynamic binarization are prior art in the field, and will not be described here.

[0043] In the Fourier transform results obtained by analyzing the horizontal and vertical frequencies in step four, as shown in Figures 5 to 8 , the width of the periodic signal and the phase information of the periodic signal in the current image can be extracted; using these two kinds of information, the actual position of the periodic background in the image can be marked, and the invasion position relationship between the defect 1 and the liquid crystal grid 3 can be identified by combining the defect position and the actual position of the periodic background, thereby assisting in judging the defect severity level. As shown in Figure 12The position of the defect 1 white spot is at the conductor 2, which belongs to the serious defect level.

[0044] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method of adaptive periodic background subtraction based on Fourier analysis, characterized in that, It comprises the following steps: Step one, image input, input the original image with defects (1) to the algorithm module; Step two, horizontal and vertical frequency division, respectively calculate the average value of each row and each column of the image, and obtain the horizontal frequency graph with only horizontal frequency and the vertical frequency graph with only vertical frequency; Step three, respectively perform inverse color, enhancement and normalization preprocessing operations on the horizontal frequency graph and the vertical frequency graph to obtain the horizontal frequency preprocessing graph and the vertical frequency preprocessing graph; Step four, Fourier transform is performed on the images of step three, Fourier transform is performed on the horizontal frequency preprocessing graph to obtain the horizontal frequency spectrum response graph and the horizontal frequency phase response graph; Fourier transform is performed on the vertical frequency preprocessing graph to obtain the vertical frequency spectrum response graph and the vertical frequency phase response graph; Step five, calculate the high amplitude phase difference of the images of step four; Step six, obtain the horizontal and vertical phase; Step seven, high response point elimination; Step eight, inverse Fourier transform, inverse Fourier transform is performed on the separated spectrum response graph and phase response graph in step seven; Step nine, feature recognition; In step six, the horizontal and vertical phase is obtained by calculating the positions of a plurality of frequency maximum values of the horizontal frequency spectrum response graph, then calculating the corresponding horizontal frequency phase response positions, and then obtaining all the periodic characteristic information in the horizontal direction, i.e. the horizontal frequency fundamental frequency position; the positions of a plurality of frequency maximum values of the vertical frequency spectrum response graph are calculated, then the corresponding vertical frequency phase response positions are calculated, and then the periodic characteristic information in the vertical direction is obtained, i.e. the vertical frequency fundamental frequency position; In step seven, the high response point elimination includes performing two-dimensional bidirectional Fourier transform on the original image with defects (1) to obtain the original image spectrum response graph and the original image phase response graph, and eliminating the horizontal frequency and vertical frequency corresponding fundamental frequency positions on the original image spectrum response graph and the original image phase response graph to separate the periodic and non-periodic characteristics in the horizontal and vertical directions.

2. The adaptive periodic background elimination method based on Fourier analysis according to claim 1, wherein the preprocessing operation in step three, the inverse color preprocessing operation is to take the important dark lines as positive signals for Fourier analysis; the enhancement preprocessing operation is to adjust the contrast of the image to make the step of the positive signal more obvious; and the normalization preprocessing operation is to linearly scale the maximum value and the minimum value of the image to the top and bottom of the color gamut; the purposes of the above inverse color, enhancement and normalization preprocessing operations are to make the results after Fourier transform clearer and easier to identify the maximum response area.

3. The adaptive periodic background elimination method based on Fourier analysis according to claim 1, wherein the feature recognition in step nine, the defects (1) can be clearly seen in the result after the periodic characteristics are eliminated, and the traditional image algorithm of threshold segmentation or dynamic binarization can be used to extract the defect position.

4. The adaptive periodic background elimination method based on Fourier analysis according to claim 1, wherein the Fourier transform results obtained by analyzing the horizontal and vertical frequencies in step four can extract the width of the periodic signal and the phase information of the periodic signal in the current image. ​ ​ ​ Using the two kinds of information, the actual position of the periodic background in the image can be marked, and the invasion position relationship between the defect (1) and the liquid crystal grid (3) can be identified by combining the position of the defect (1) and the actual position of the periodic background, thereby assisting in judging the defect severity level.

Images